CSCI_5980_Presentati.. - University of Minnesota Twin Cities

Report
CrimeStat
GROUP 10
NORINE WILCZEK & BRAD JOHNSTON
CSCI 5980
DECEMBER 4, 2012
Organization
• About CrimeStat
• CrimeStat analysis tools
• Problem & importance
• Data
• Challenges
• Tools & methods used
• Processes & map outputs
• Limitations
• Contributions to computers and society
About CrimeStat
• Analysis tool used in
•
•
•
Law enforcement
Public health
The environment
• Free Download from National Institute of Justice
• http://www.icpsr.umich.edu/CrimeStat/
CrimeStat Continued
• Spatial statistics program
• Windows based
• Purpose: provide supplemental statistical tools to aid law
enforcement agencies and criminal justice researchers in
crime mapping
• Uses GIS shapefiles to perform object-based analysis
• Primary file
•
Incident locations with X,Y coordinate system
• Secondary file
•
for comparison
• Reference file
•
Grid overlay for measurement, used for model interaction of 2
points
CrimeStat Analysis Options
• Spatial Description
•
Distance Analysis
•
•
•
Spatial Autocorrelation
•
•
Nearest neighbor, linear
nearest neighbor, or Ripley’s
K statistic distance between
incidents
Calculates distance between
incidents from 2 files and
places on a grid
Stats for describing amount
of spatial autocorrelation
between incidents
Spatial Distribution
•
•
Mean center
Center of minimum distance
•
Hotspot Analysis
•
•
•
Mode, fuzzy mode,
hierarchical nearest
neighbor clustering
Risk-adjusted nearest
neighbor hierarchical
clustering -> ellipses or
convex hull output
Spatial and Temporal
Analysis of Crime (STAC), Kmeans cluster, Anselin’s
local Moran, Getis-Ord local
G statistics -> ellipses or
convex hulls
More CrimeStat Analysis Options
• Spatial Modeling
•
Regression modeling
• Crime Travel Demand Models
•
• Analyzes relationship between a
dependent variable and one or
more independent variables
•
Journey to Crime
• Predict number of crimes in each
zone (origins) and (destinations)
•
•
Space-time Analysis
each zone to every other zone
using gravity model
•
zone to zone using function that
approximates one mode relative to
other modes.
(serial offender data)
Interpolation
• Single variable kernel density
• dual-variable kernel density
(comparing to baseline)
Mode Split
• Split predicted number of trips for
• Clustering in time and space
•
Trip Distribution
• 2nd stage, distributes trips from
• Serial offender data – likely
location based on distribution of
incidents and travel behavior
Trip Generation
•
Network Assignment
• Shortest path algorithm predicts
trips from each zone to other zone
(likely path). Requires travel
network (transits & one way
streets, roads, etc)
Problem & Importance
• Problem
•
•
•
Crime occurs globally
Statistical analysis is necessary
Patterns, trends, high crime areas, potential re-offending
predictions
• Importance
•
•
Response
Prevention
•
•
•
•
•
Crime
Injuries
Death
Utilize resources
Mitigation of economic losses
•
Lost/Recovered property
Data
• University of Minnesota
Police Department
•
•
•
9/2011 – 9/2012
September 2011
• (all crimes)
Theft from building
•
•
(9/11 – 9/12)
Bicycle thefts
Challenges
• Data
•
•
Process through a GIS
View results with a GIS
•
.shp, .dbf (uses and produces shapefiles, not feature classes)
• Clean up received data
•
•
Time/Date field
City, state, zip field → Google
•
Proper geo-coding in ArcMap
Tools & Methods Used
• Spatial Distribution Tool
• Distance Analysis Tool
• Hotspot Analysis
Spatial Distribution Tool
• Mean & median center, center of minimum distance
• Standard deviation
•
Half of crimes in a cluster will be within one standard
deviation ellipse of the mean center, around 90% will be found
within two standard deviation ellipses
• Forecasting: identifying where crime is likely to
occur
Result
Map
Result: Map
Distance Analysis Tool
• Distance Analysis 1
•
Point pattern of clustering and dispersion
•
•
Distances between the points and reference locations as indicator
(distance based tests)
Number of points in a given area for basis of test statistics
•
•
If distance is smaller than what it would be under complete spatial
randomness, it suggests clustering
If distance tends to be larger, then it suggests dispersion
Result: Chart
Nearest neighbor analysis:
-------------------------Sample size........: 26
107.64 sq ft
Measurement type...: Direct
0.00000 sq mi
Start time.........: 03:49:10 PM, 11/05/2012
Mean Random Distance ............: 0.31 m, 1.02 ft, 0.00019 mi
Mean Nearest Neighbor Distance ..: 109.91 m, 360.58 ft,
Mean Dispersed Distance .........: 0.67 m, 2.19 ft, 0.00041 mi
0.06829 mi
Nearest Neighbor Index ..........: 354.4364
Standard Dev of Nearest
Neighbor Distance ...............: 183.07 m, 600.61 ft, 0.11375 mi
Minimum Distance ................: 0.00 m, 0.00 ft, 0.00000 mi
Maximum Distance ................: 2611.89 m, 8569.19 ft, 1.62295
mi
Standard Error ..................: 0.03 m, 0.10 ft, 0.00002 mi
Test Statistic (Z) ..............: 3447.6946
p-value (one tail) ..............: 0.0001
p-value (two tail) ..............: 0.0001
Mean Nearest
Based on Bounding Rectangle:
Area ............................: 3808465.65 sq m
40993983.06 sq ft
1.47046 sq mi
Mean Random Distance ............: 191.36 m, 627.83 ft, 0.11891
mi
Mean Dispersed Distance .........: 411.25 m, 1349.25 ft, 0.25554
mi
Nearest Neighbor Index ..........: 0.5743
Standard Error ..................: 19.62 m, 64.36 ft, 0.01219 mi
Test Statistic (Z) ..............: -4.1523
p-value (one tail) ..............: 0.0001
p-value (two tail) ..............: 0.0001
Based on User Input Area:
Area ............................: 10.00 sq m
Order
Index
Expected Nearest Nearest
Neighbor Distance (m) Neighbor Distance (m) Neighbor
*****
********************* *********************
**************
1
109.9061
0.3101
354.43636
.5743
Cluster
Result Map
Bike Thefts
2011-2012
Hotspot Analysis
•
•
Hotspot: dense area of incidents, in
this case a spatial concentration of
crime
"Geographic area representing a
small percentage of the study area
which contains a high percentage of
the studied phenomena"
• Spatial Description
•
Hotspot Analysis I
• Fuzzy Mode
• identifies the geographic
coordinates, plus a userspecified surrounding radius,
with the highest number of
incidents
•
Nearest neighbor Hierarchical
Spatial Clustering (Nnh)
•
•
Interpolation method
Minimum points per cluster
• Results:
•
•
•
7 NNH clusters
10 or more Building Thefts within
1500 sq. meter area
Calculates mean X,Y of ellipses in
the output table
Mode vs. Fuzzy Mode
Mode
Fuzzy Mode
•
• Hotspot
Analysis
Result
Map
•
•
Theft from
Buildings
•
2011-2012
Kernel Density Estimation
• Most popular type of map in
•
•
•
•
•
crime analysis
Generalized over larger areas
(compared to Hotspot)
Interpolation method
Creates “risk areas”
Kernel size and weight
determined by user, smoothed
(linear relationship) throughout
kernel
Multiple points at one location,
kernels aggregate to total in grid
cell
Kernel
Density
Estimation
Analysis
Result
Map
Theft from
Buildings
2011-2012
Nearest
Neighbor
Hiearchical
& Kernel
Density
Estimation
*Nearest neighbor
clusters and kernel
density estimation
analysis overlay
*Mondale Hall & Carlson
*Coffman Mem Union
*Walter, Appleby, &
Johnston Hall area
*Rec Center
Limitations
• CrimeStat uses data with latitude and longitude
•
•
•
Does not pick up “on the fly”
Spatial references need to match
Our data missing X,Y column
• Add XY Data tool in ArcMap
•
Experimented with and added in XY data for use in CrimeStat
• Size of geographic region
•
•
CrimeStat useful for larger areas (than U of M campuses)
Clusters would show up in a city or regional level where areas have
crime that is less likely to occur (including stats of bike ridership,
socioeconomic conditions)
Contributions to Computer & Society
• Analysis tool for Law Enforcement, Public Health,
and the Environment
• Visual analysis vs. statistical analysis
• Benefits of CrimeStat:
•
Calculates spatial statistics, which can calculate correlations
between geographic variables and detect subtle changes in
geography of a pattern over time that they eyes do not see
• Law enforcement resource allocation
• Faster response time
• Citizen awareness
End of Presentation
Questions ?

similar documents